基于无人机高分辨率影像的油松新造林健康树冠提取  

Extraction of Healthy Canopy of New Afforestation for Pinus tabulaeformis Based on UAV High-Resolution Image

作　　者：郭旭展 陈巧[1,3] 张晓芳[1,3] 洪亮 尤媛媛 唐守正 符利勇 Guo Xuzhan;Chen Qiao;Zhang Xiaofang;Hong Liang;You Yuanyuan;Tang Shouzheng;Fu Liyong(Research Institule of Forest Resource Information Techniques,CAF Beijing 100091;College of Computer and Information Technology,Xinyang Normal University Xinyang 464000;Key Laboratory of Forest Management and Growth Modeling,National Forestry and Grassland Administration Beijing 100091;College of Mathematics and Statistics,Xinyang Normal University Xinyang 464000;Forestry and Grassland Bureau of Chongli District,Zhangjiakou City,Hebei Province Zhangjiakou 075000)

机构地区：[1]中国林业科学研究院资源信息研究所,北京100091 [2]信阳师范学院计算机与信息技术学院,信阳464000 [3]国家林业和草原局森林经营与生长模拟重点实验室,北京100091 [4]信阳师范学院数学与统计学院,信阳464000 [5]河北省张家口市崇礼区林业和草原局,张家口075000

出　　处：《林业科学》2022年第10期111-120,共10页Scientia Silvae Sinicae

基　　金：张家口市崇礼区森林防火综合体系建设无人机巡护监测系统(DA2020001);国家自然科学基金面上项目(31971653)。

摘　　要：【目的】基于新造林健康树冠的光谱特征和空间交错情况,探讨复杂地面植被条件下健康树冠的光谱增强方式和多尺度分割阈值,为造林核查的日常监测工作提供技术支撑。【方法】以冬奥核心区新造林地无人机航拍影像为试验数据,首先,基于健康树冠与其他干扰地物的不同颜色特征,采用同态滤波增强影像并使用ExG光谱指数进行变换;然后,采用最大类间方差方法得到二值图像,并使用多尺度形态学滤波方法进行分割并融合分割结果,以分割交错的树冠区域对应提取原始图像中可能的健康树冠区域;最后,基于颜色向量、灰度共生矩阵和局部二值模式共同构建的特征向量,采用随机森林识别提取区域,从而检测图像中的健康树冠。【结果】基于光谱指数变换、多尺度形态学滤波方法能够有效分割交错连续的树冠区域,排除与健康树冠颜色相近的地物干扰,较为准确提取出可能为树冠的区域。采用该方法对不同造林密度、光照条件下的17幅无人机正射图像进行试验,使用目视解译方式标记出树冠中心,运用精确度、召回率和F1分数3个评价指标对随机森林和支持向量机的识别效果进行定量对比分析,结果表明,多尺度形态学滤波方法可提取96.78%的树冠,随机森林的F1分数高于97%,而支持向量机的召回率显著低于随机森林。【结论】基于光谱指数变换和多尺度形态学滤波的树冠提取方法能够对健康树冠进行快速、准确提取,有效完成造林核查。【Objective】Based on the spectral characteristics and spacial interlacing of healthy tree crowns in new afforestation,the spectral enhancement method and multi-scale segmentation thresholds of healthy tree crowns under complex ground vegetation conditions were discussed to provide technical support for the daily monitoring of afforestation verification.【Method】The UAV images of newly planted trees in the core area of the Winter Olympics were chosen as experimental data.Firstly,based on the different color characteristics of healthy tree crowns and other disturbances,the images were enhanced by homomorphic filtering and transformed by ExG spectral index.Then,the Otsu method was used to obtain the binary image,and the multi-scale morphological filtering method was used for segmentation and fusion to segment the interlaced crown areas,correspondingly extract the possible healthy crown areas in the original image.Finally,based on the feature vector constructed by the color vector,the GLCM and the LBP,the random forest was used to classify the extracted area to detect the healthy tree crowns in the image.【Result】The method based on spectral index transformation and multi-scale morphological filtering was able to effectively segment the interlaced and continuous crown areas,exclude other interference objects those were similar in color to healthy trees and accurately extract the areas those might be crowns.The 17 UAV orthophoto images with varying stand densities and lighting conditions were tested,and the crown centers were marked by visual interpretation.Furthermore,the three evaluation indexes:precision,recall and F1 score were used to quantitatively compare and analyze the recognition effects of random forest and SVM.The experimental result showed that 96.78%of the crowns were extracted using the multi-scale morphological filtering method,and the F1 score of random forest was higher than 97%,while the recall of support vector machine was significantly lower than that of random forest.【Conclusion】Our

关 键 词：无人机影像 同态滤波 光谱指数 形态学滤波 随机森林 

分 类 号：S758[农业科学—森林经理学]